With the rapid development of hydropower engineering in China, the long-term operational safety of high arch dams, as critical and complex structures, has emerged as a pressing scientific and engineering challenge. This study investigates the evolutionary patterns and variation characteristics of multi-source monitoring data for high arch dams, employing unsupervised anomaly detection algorithms to validate and cleanse abnormal data, thereby ensuring the reliability and validity of monitoring datasets. Subsequently, a multi-point statistical model based on stepwise regression analysis is established to fit and characterize the dynamic behavior of typical monitoring points. This approach elucidates the dominant influencing factors and their dynamic variation mechanisms governing dam deformation. The proposed methodology provides a robust theoretical foundation and technical framework for long-term structural health monitoring and early warning systems of high arch dams, effectively addressing the inherent complexities of spatial deformation and operational uncertainties.